Analysis of Genomic Sequences of 95 Papillomavirus Types: Uniting Typing, Phylogeny, and Taxonomy

Transcription

1 JOURNAL OF VIROLOGY, May 1995, p Vol. 69, No X/95/$ Copyright 1995, American Society for Microbiology Analysis of Genomic Sequences of 95 Papillomavirus Types: Uniting Typing, Phylogeny, and Taxonomy SHIH-YEN CHAN, 1 HAJO DELIUS, 2 AARON L. HALPERN, 3 AND HANS-ULRICH BERNARD 1 * Laboratory for Papillomavirus Biology, Institute of Molecular and Cell Biology, National University of Singapore, Singapore 0511, Republic of Singapore 1 ; German Cancer Research Center, Heidelberg, Germany 2 ; and Theoretical Biology and Biophysics, Theoretical Division, Los Alamos National Laboratory, Los Alamos, New Mexico Received 26 September 1994/Accepted 16 February 1995 Our aim was to study the phylogenetic relationships of all known papillomaviruses (PVs) and the possibility of establishing a supratype taxonomic classification based on this information. Of the many detectably homologous segments present in PV genomes, a 291-bp segment of the L1 gene is notable because it is flanked by the MY09 and MY11 consensus primers and contains highly conserved amino acid residues which simplify sequence alignment. We determined the MY09-MY11 sequences of human PV type 20 (HPV-20), HPV-21, HPV-22, HPV-23, HPV-24, HPV-36, HPV-37, HPV-38, HPV-48, HPV-50, HPV-60, HPV-70, HPV-72, HPV-73, ovine (sheep) PV, bovine PV type 3 (BPV-3), BPV-5, and BPV-6 and created a database which now encompasses HPV-1 to HPV-70, HPV-72, HPV-73, seven yet untyped HPV genomes, and 15 animal PV types. Three additional animal PVs were analyzed on the basis of other sequence data. We constructed phylogenies based on partial L1 and E6 gene sequences and distinguished five major clades that we call supergroups. One of them unites 54 genital PV types, which can be further divided into eleven groups. The second supergroup has 24 types and unites most PVs that are typically found in epidermodysplasia verruciformis patients but also includes several types typical of other cutaneous lesions, like HPV-4. The third supergroup unites the six known ungulate fibropapillomaviruses, the fourth includes the cutaneous ungulate PVs BPV-3, BPV-4, and BPV-6, and the fifth includes HPV-1, HPV-41, HPV-63, the canine oral PV, and the cottontail rabbit PV. The chaffinch PV and two rodent PVs, Micromys minutus PV and Mastomys natalensis PV, are left ungrouped because of the relative isolation of each of their lineages. Within most supergroups, groups formed on the basis of cladistic principles unite phenotypically similar PV types. We discuss the basis of our classification, the concept of the PV type, speciation, PV-host evolution, and estimates of their rates of evolution. Downloaded from During the last decade, papillomaviruses (PVs) have attracted increasing scientific attention, as they are quantitatively the most important group of viruses associated with benign and malignant neoplasia in humans (40, 73). Currently, more than 70 different human PV (HPV) types are known (15), and additional evidence from partial sequences indicates the existence of a further 13 which would qualify as novel HPV types (2, 5). Other PV types have also been identified in mammals and birds (61). The large number of PV types and poorly defined lower- (e.g., subtypes) and higher-order classifications (e.g., mucosal HPVs) have become a potential source of confusion, and it seems desirable to investigate the possibility of giving them some internal taxonomic organization. Before this is done, it is necessary to have a clear definition of the objects one is classifying and reliable hypotheses about their phylogeny. PV types are the objects to be classified, and they are defined by genomic properties rather than serology; thus, the term serotype is not used. The original definition was by DNA hybridization criteria (9), but this method had a serious limitation in that the values determined for genomic similarities were frequently very different from and inconsistently related to nucleotide sequence similarities. In fact, PVs which show little or no relatedness by hybridization could still have a greater than 50% nucleotide similarity. This is due to the scattered * Corresponding author. Mailing address: Laboratory for Papillomavirus Biology, Institute of Molecular and Cell Biology, National University of Singapore, Singapore 0511, Republic of Singapore. Fax: (65) distribution of short sequence homologies. To overcome this inconsistency, an HPV genome is now defined as a new HPV type when it shows a more than 10% dissimilarity in the combined nucleotide sequences of the E6, E7, and L1 genes when compared with those of any previously known type (15). This requires the sequencing of about 2.4 kb, or roughly one-third, of the genome of all new isolates and then sequence alignment and the calculation of dissimilarity according to certain operational criteria. Phylogenetic research uses homologous features such as nucleotide sequences to reconstruct evolutionary relationships which are graphically represented as trees. As a result of large systematic sequencing efforts (for a review, see reference 13), the PVs have become the best-known DNA virus family and an important resource for the study of virus evolution (4, 7, 31, 50, 65). Since the first extensive phylogenetic studies of PVs were done (7, 65), there has been a doubling of the number of PV types for which complete or partial sequence information is known. There are indications that the rate at which new genital HPV types are detected is slowing down (5, 66), while only small efforts to detect novel PVs in animals are presently made. However, many new HPV genomes related to the epidermodysplasia verruciformis (EV) HPVs have recently been found in skin cancers from renal transplant patients (2, 57). Because of this increase in the number of known sequences, the advent of fresh analytical approaches, and the need for taxonomic investigation, we felt it would be helpful to undertake a revision of PV phylogeny. Given the objects to be classified and their phylogeny, there are broadly three main taxonomic approaches one can take. on June 19, 2018 by guest 3074

2 VOL. 69, 1995 PV PHYLOGENY AND TAXONOMIC CLASSIFICATION 3075 The differences arise mainly from the different philosophical perceptions of what a taxonomic scheme should represent and what qualities it should possess. The cladistic approach uses evolutionary relationships deduced from shared derived characters, and it establishes taxonomic groups that must be monophyletic groups (clades) (60, 70). Pheneticists use an overall similarity measure (or its complement, the dissimilarity) of reliably measured phenotypic characters (59). Finally, evolutionary systematists use a combination of both criteria (43), differing from that of cladists in that they also accept nonmonophyletic groups as taxonomic groups. Genetic sequence data may be used by all the schools because the DNA sequence is both inherited material suitable for phylogenetic evaluation and also a phenotypic feature that can be determined reliably. There exist other phenotypic features such as virion morphology, protein structure, disease pathology, and immune responses that could conceivably enter into a taxonomic scheme. However, many of these either are poorly understood, are not distinctive enough to provide a foundation for a taxonomy, or may also have resulted from sharing an ancestral trait or from independent evolution of the trait. They, however, may play an adjunct role, and as knowledge of these features increases, it will be interesting to see how well they correlate to the existing groups. We adopted the cladistic methodology, as it is a particularly consistent and useful approach for classifying PVs. MATERIALS AND METHODS Origin of PV clones and sequences. Original references to the isolation of all the HPV types numbered 1 to 69 are contained in three reviews (13, 14, 66). These reviews and reference 48 also cite all publications of the complete genomic nucleotide sequences of 43 HPV types. We cite only those concerning specific statements made in this study. The sequences of all HPV types and all animal PVs not mentioned separately below were obtained from GenBank (Release 83.0) and have been recently published as part of a compendium (48). This compendium also contains partial sequences of seven HPV genomes which will probably attain type status after complete isolation and DNA sequencing. These seven clones were identified during several major epidemiological studies summarized in reference 5, and their codes (CP, IS, LV, and MM) refer to the respective epidemiological studies. Three additional genomes found during this latter study were assumed to be novel HPV types but have since then become identified as variants of HPV-70 (CP141), HPV-72 (CP4173-LVX100), and HPV-73 (MM9) (68). Since the genomic sequences of the reference clones of these three types were not yet available, we used the sequences of these three variants to represent these types. The L1 sequences of HPV-20, HPV-21, HPV-22, HPV-23, HPV-24, HPV-36, HPV-37, HPV-38, HPV-48, HPV-50, HPV-60, and ovine PV were determined by one of us (H.D.) during an ongoing project to establish the complete genomic sequences of most PVs. The sequences of bovine PV type 3 (BPV-3) and BPV-6 were obtained with sequencing primers modeled after the nucleotide sequence of their presumed nearest relative, BPV-4 (34), and that of BPV-5 was obtained by amplifying it with PCR primers modeled after BPV-1. These sequences are available by anonymous file transfer protocol from the HPV database (atlas@lanl.gov). The clones of HPV-20, HPV-21, HPV-22, HPV-23, HPV-24, HPV-36, HPV-37, HPV-38, HPV-48, HPV-50, and HPV-60 were obtained from E.-M. de Villiers at the Reference Center for Human Pathogenic Papillomaviruses, Heidelberg, Germany. The clone of ovine PV was supplied by Phillip Baird, Sydney, Australia, and the clones of BPV-3, BPV-5, and BPV-6 DNA were furnished by Saveria M. Campo, Glasgow, United Kingdom. Data analysis. Sequence alignments, distances, and manipulations were done with the Genetics Computer Group Sequence Analysis Software Package version UNIX (23) and the Multiple Aligned Sequence Editor (20). The pairwise simple distance (dissimilarity) was determined on aligned sequences, excluding gaps, and is the proportion of nucleotides that differed for a pair of sequences. Linear correlation analysis, with the best-fit line additionally constrained to pass through the origin, was performed with the FIT and REGRESS programs in Mathematica version 2.2 (Wolfram Inc., Champaign, Ill.). The difference between constrained and unconstrained fits was negligible. Distance matrix and maximum-likelihood phylogenies and distance matrix bootstrap analyses were done with the Phylogeny Inference Package version 3.5 (21, 22) and the fastdnaml version 1.0 program (49). Maximum and weighted parsimony analyses were conducted with the Phylogenetic Analysis Using Parsimony package version (62). The weighted parsimony analyses (29) used an inverse frequency substitution matrix generated with the help of MacClade version (41). Multiple Aligned Sequence Editor, Genetics Computer Group Software, and Phylogeny Inference Package runs were performed on a Silicon Graphics server running IRIX V.4 and a Tatung Sparc 5 model 85 running Solaris 2.3. Phylogenetic Analysis Using Parsimony and MacClade were run on a Macintosh Quadra 800 running System Software 7.1. Mathematica was run on an IBM PC-compatible 486 running OS/ RESULTS Alignment of homologous partial L1 sequences of 92 PV types. Strict amino acid conservation allows the alignment of several genomic segments among all presently known PV genomes. Specifically, good alignment is possible for the central portion of the E6 gene (10, 67), the DNA binding and the transcription activation domains of the E2 gene (24), and several gene segments of the E1, L2, and L1 genes (7, 48, 64). Alignment is not possible for several segments that have accumulated extensive substitutions, deletions, and insertions, and such segments include the overlap of the E2 hinge region and the E4 gene (16) and the E5 gene (1). The conserved segments are useful for phylogenetic analysis, provided that they are large enough to be sufficiently informative. In previous studies of two 152-bp and 132-bp segments of the E1 and L1 genes, we found similar relationships among 48 PV types irrespective of the nature of the evaluated genomic segment (7). Most of these affinities were in agreement with those found during another study of a 384-bp segment of the E6 gene (65). Figure 1 shows the alignment of the amino acid sequences that correspond to part of the L1 gene of 92 PV types and presently represents the most extensive database for any PV segment. It resulted from the extension of research on the detectability of unknown genital HPV genomes (5) with the PCR primer pair MY09-MY11 (42). These primers amplify a segment of approximately 460 bp of the L1 gene of most genital HPV types. A 291-bp subsegment of this amplimer can be confidently aligned because of highly conserved amino acid residues which are indicated in Fig. 1. This particular subsegment excludes sequences corresponding to those of the MY09-MY11 primers and the 5 portion of the segment which is of low conservation, with numerous small insertions and deletions (5). In the following, we refer to these sequences as the 291-bp L1 segments. The database did not include sequences for the colobus monkey PV (CgPV) (53), the only avian (chaffinch) PV (FPV) (46) analyzed so far, and the Micromys minutus (mouse) PV (MmPV) (67), because the relevant L1 sequences and the cloned genomes were not available. However, in addition to all the types whose full sequences are published, we did evaluate about 30 partially sequenced HPVs; BPV-3, BPV-4, and BPV-6, which do not have E6 genes (34); and animal PVs that have not attracted much sequencing effort. The 291-bp L1 segment and E6 gene contain sufficient sequence information for typing. Given the large amount of data, the ready accessibility of the 291-bp L1 segment by PCR, and its utility for phylogenetic studies, we studied how it compared with the combined E6-E7-L1 segment for typing purposes. To the extent that the 291-bp L1 segment could be used to reliably replicate the definition of types determined with the full E6, E7, and L1 sequences, this would give us confidence that the typing of novel sequences with the 291-bp segment would yield the same result as would typing based on the current standard. A potential concern in this regard, and in using the 291-bp segment for phylogenetic purposes, was that such a small wellconserved segment might lack the resolving power required for these analyses. Using the alignments from the HPV database compendium (48), we constructed a combined alignment of E6, E7, and L1 for all types for which the sequences of all three genes were

3 3076 CHAN ET AL. J. VIROL. FIG. 1. Alignment of the amino acid sequences corresponding to the highly conserved 291-bp L1 segments of 92 PV types. The upper part of the figure indicates the genomic position of this sequence in the L1 gene as well as those of the MY09-MY11 PCR primer pair. The solid arrows indicate amino acid residues that are strictly conserved in all PVs. These and many other highly conserved amino acid residues are useful for aligning the sequences and indicating the PV origin of partial genomic segments (5). Striped arrows indicate amino acid residues that strictly covary between genital and EV HPVs, i.e., the sequence MXYXH of genital PVs becomes LXQXN in EV HPVs. HPV-68, HPV-70, HPV-72, and HPV-73 sequences may have slight nucleotide sequence differences from the reference isolates. Abbreviations: LCR, long control region; BPV, bovine PV; CgPV, colobus monkey PV; COPV, canine oral PV; CRPV, cottontail rabbit PV; DPV, deer PV; EPV, European elk PV; FPV, chaffinch (avian) PV; HPV, human PV; MmPV, Micromys minutus (mouse) PV; MnPV, Mastomys natalensis (South African mouse) PV; OvPV, ovine (sheep) PV; PCPV, pigmy chimpanzee PV; RhPV, rhesus monkey PV. Note the absence of HPV-46, which is now considered to be a subtype of HPV-20 (15). FIG. 2. Linear correlation analyses of all available 291-bp L1 (A) and 399-bp E6 distances against the combined E6, E7, and L1 combined gene distances, the latter being derived from published alignments (48). Over 2,000 datum points were plotted for each anlaysis. The best-fit line was constrained to pass through the origin by adjusting the model parameters within Mathematica. available. After removing all columns from the combined alignment which contained gaps in one or more sequences, we calculated dissimilarities for each pair of sequences as the number of mismatches divided by the number of aligned positions compared (e.g., 25 mismatches out of 200 positions 12.5%). After doing the same for the 291-bp L1 segment and for a 399-bp E6 segment (see below), we attempted a linear correlation analysis. The results plotted in Fig. 2 indicate that both the 291-bp L1 and the 399-bp E6 segment distances are highly linearly correlated with those of the E6-E7-L1 sequences used for typing (r 0.99 in both cases), with gradients of 0.92 and 1.28, respectively. The gradients suggest that the 291-bp L1 segment sequence is on the average more conserved and that the 399-bp E6 segment sequence is less conserved than the E6-E7-L1 sequences. Thus, the current 10% criterion for typing is roughly equivalent to 9.2% and 12.8% criteria, respectively, if these other segments are used. We further checked the distances between several of the most closely related pairs of HPV types to see if using the smaller 291-bp L1 distances would frequently lead to a disagreement with the current standard on the type status. Table 1 shows that even the most closely related types have distances that significantly exceed for the 291-bp L1 segment, and therefore we conclude that the shorter 291-bp L1 distances do not normally lead to errors about type status. However, there are two pairs with clearly lower 291-bp distances, HPV-34 and HPV-64 (0.050) and HPV-44 and HPV-55 (0.057). The preliminary genomic sequence data for HPV-64 (15) in fact suggest that the HPV-34 HPV-64 pair may not be separate types. The verdict on the HPV-44 HPV-55 pair awaits more sequence information.

4 VOL. 69, 1995 PV PHYLOGENY AND TAXONOMIC CLASSIFICATION 3077 TABLE 1. Comparison of pairwise distances for 10 very closely related HPV types HPV pair E6-E7-L1 distance 291-bp L1 distance HPV-2 HPV HPV-2 HPV HPV-3 HPV HPV-4 HPV HPV-5 HPV HPV-6 HPV HPV-7 HPV HPV-11 HPV HPV-18 HPV HPV-19 HPV The phylogeny of 92 PV types as determined on the basis of partial L1 sequences. Figure 3 shows a maximum-likelihood phylogeny of 92 PV types as determined on the basis of nucleotide sequences of the 291-bp L1 segments (Fig. 1). Neighborjoining analysis gave results very similar to those given by maximum likelihood, regardless of distance corrections (results not shown). A maximum-parsimony analysis led to a large number of equally parsimonious trees. This finding, in conjunction with the highly divergent character of the sequences, suggests that there is a lot of noise due to back mutation and homoplasy and prompted the use of the weighted parsimony approach, which led to the tree shown in Fig. 4. In both trees, the same two large branches that have been observed before (5, 7, 8, 65) are clearly present. One unites HPV types associated with genital lesions (genital HPVs, referred to in the following as supergroup A), and the other unites HPV types associated with EV (supergroup B). On FIG. 3. A phylogenetic tree of 92 PV types based on a maximum-likelihood evaluation of the 291-bp L1 segment. In addition, FPV, CgPV, and MmPV are included in this tree, but their relationships are indicated by dashed branches, because they were derived from different sequence data. For abbreviations, see the legend to Fig. 1. FIG. 4. A phylogenetic tree of 92 PV types based on a weighted parsimony evaluation of the 291-bp L1 segment. Branch lengths are proportional to the numbers of weighted steps between reconstructed bifurcations of lineages. An initial unweighted analysis involving 10 random sequence addition replicates and nearest neighbor interchange searching yielded 217 equally parsimonious trees of 4,021 steps. An inverse-frequency substitution weight matrix was generated with MacClade as follows. The average number over all 217 trees of each of the 12 possible nucleotide substitutions was calculated, counting only unambiguously reconstructed changes. Weights for each type of substitution were then calculated as the total number of changes of all types over the number of changes of that particular type. Weighted parsimony analyses were performed with these weights. Nearest neighbor interchange replicates (n 75) yielded a single best tree with a weighted length of 35,564. Additional ad hoc searches involving subtree pruning-regrafting on user-defined starting trees yielded a better tree with a weighted length of 35,536, presented here. The abbreviations are as defined in the legend to Fig. 1. branches positioned between them are most animal PVs and three HPVs associated with cutaneous lesions, namely, HPV-1, HPV-41, and HPV-63. One of these branches unites all fibropapillomaviruses from ungulates, and we call this supergroup C. A fourth branch unites BPV-3, BPV-4, and BPV-6 (supergroup D), and the fifth unites HPV-1, HPV-41, HPV-63, canine oral PV, and cottontail rabbit PV (supergroup E). Table 2 summarizes the PV types within these supergroups and further proposes groups within these supergroups formed of types with relatively close relationships. Details of the rationale behind this classification will be discussed below. Both phylogenies are very similar, but we note the following differences (see Table 2 for the group identities). (i) There are internal topology differences for supergroup A, e.g., the various positions of groups A1, A3, A5, and A6. (ii) There are internal differences for group B1. (iii) Supergroups C and D (all BPVs) together form a monophyletic group in Fig. 3 but not in Fig. 4 or in a bootstrapped consensus neighbor-joining tree (data not shown). (iv) Supergroup E, although apparently a clade, has members that are very distantly related. Other causes of uncertainty arise from the large number of types, making the search for the optimal tree necessarily far

5 3078 CHAN ET AL. J. VIROL. TABLE 2. Classification of PVs on the basis of cladistic relationships PV supergroup a Supergroup A (genital HPVs) Group A1: HPV-32 and HPV-42 (100%) Group A2: HPV-3, HPV-10, HPV-28, and HPV-29 (100%) Group A3: HPV-61, HPV-62, HPV-72, CP6108, CP8304, and MM8 (94%) Group A4: HPV-2, HPV-27, and HPV-57 (100%) Group A5: HPV-26, HPV-51, HPV-69, ISO39, and MM4 (97%) Group A6: HPV-30, HPV-53, HPV-56, and HPV-66 (100%) Group A7: HPV-18, HPV-39, HPV-45, HPV-59, HPV-68, and HPV-70 (78%) Group A8: HPV-7, HPV-40, and HPV-43 (100%) Group A9: HPV-16, HPV-31, HPV-33, HPV-35, HPV-52, HPV-58, HPV-67, and RhPV (34%) Group A10: HPV-6, HPV-11, HPV-13, HPV-44, HPV-55, and PCPV (100%) Group A11: HPV-34, HPV-64, and HPV-73 (100%) Poorly resolved lineages within supergroup A: CgPV (291-bp L1 sequences unavailable), CP8061, LVX82, and HPV-54 Supergroup B (EV HPVs) Group B1: HPV-5, HPV-8, HPV-9, HPV-12, HPV-14, HPV-15, HPV-17, HPV-19, HPV-20, HPV-21, HPV-22, HPV-23, HPV-24, HPV-25, HPV-36, HPV-37, HPV-38, HPV-47, and HPV-49 (100%) Group B2: HPV-4, HPV-48, HPV-50, HPV-60, and HPV-65 (73%) Supergroup C (ungulate fibropapillomaviruses) Group C1: BPV-1 and BPV-2 (94%) Group C2: DPV, EPV, and OvPV Isolated lineage within supergroup C with likely group rank: BPV-5 Supergroup D (BPVs causing true papillomas) Group D1: BPV-3, BPV-4, and BPV-6 (86%) Supergroup E (animal and human cutaneous PVs) Group E1: HPV-1 and HPV-63 Isolated lineages within supergroup E with likely group rank: COPV, CRPV, and HPV-41 Poorly resolved lineages which likely represent taxa of supergroup rank FPV (291-bp L1 sequences unavailable) MmPV (291-bp L1 sequences unavailable) MnPV a The cladistic relationships were determined on the basis of the phylogenies depicted in Fig. 3 and 4. Supergroups A, B, C, D, and E are clades with well-recognized differences in biology. Simple distances based on the 291-bp L1 segment generally exceeded 0.37 for types from different supergroups. Within a supergroup, types from different groups generally had distances ranging from 0.30 to Types within groups generally had distances ranging from 0.10 to The large value for the upper boundary of this range is due to the large distances within group B1, which should be split into two or more smaller groups when the internal phylogeny is better known. Where applicable, the bootstrap values (100 replicates, neighbor-joining method) for the groups are indicated in parentheses as percentages. The abbreviations are the same as those defined in the legend to Fig. 1. from exhaustive, and from the further possibility of alternative weighting schemes. Nevertheless, an informal comparison of the many near-optimal trees and trees based on different weighting schemes (data not shown) does confirm that many of the groups of interest are stable. Such a comparison also reveals the more uncertain aspects of the phylogeny. Instability is particularly noteworthy for CP8061, LVX82, HPV-54, and rhesus monkey PV, although all still definitely belong to supergroup A. Neighbor-joining and bootstrapping analyses revealed a low value of confidence for group A9, which was occasionally dismembered in trees with different weighting schemes. Within supergroup B, instability was high for HPV- 49. Finally, the inclusion of Mastomys natalensis PV within either supergroup C, D, or E was an alternative to a placement on a separate lineage. We included in Fig. 3 the relative positions of FPV, CgPV, and MmPV. The 291-bp L1 sequences of these viruses were not known, and their relationships with the 92 PV types listed in Fig. 1 were derived from an E1 database in the case of CgPV (7), from E6 sequences in the case of MmPV (Fig. 4) (67), and from partial E1 and L1 sequences in the case of FPV (46). The phylogeny of 57 PV types as determined on the basis of a 399-bp E6 segment. To compare our findings with those of a previous study of E6 gene sequences (65), we used the same segment of the expanded HPV database (13) to construct phylogenies. We refer to these data as the 399-bp E6 segment, because addition of some PV types required the assumption of insertions in the original 384-bp alignment. Figure 5 shows a neighbor-joining phylogeny done with these E6 sequences. It shows all previously sequenced HPV types in positions similar to those that have been previously published (65) and displays both the stable features and instabilities mentioned before, such as those for supergroup A, the distinction between group B1 and B2, and supergroup C. Supergroup D is missing, as these BPV types do not have E6 genes. On the basis of E6 sequences, supergroup E is very diverse. Cottontail rabbit PV was excluded from supergroup E because its E6 gene could not be confidently aligned. The overall similarities between the E6- and L1-based phylogenies support previous findings that PV gene trees are a good reflection of trees of PV types because of the apparent absence of intertype recombinations. Distances are inadequate as an operational tool for higher classification. The proliferation of PV types over the past years has led to the need for more general taxonomic categories. Given the success of the distance criteria in operationally defining types, it is tempting to assume that additional levels of

6 VOL. 69, 1995 PV PHYLOGENY AND TAXONOMIC CLASSIFICATION 3079 FIG. 5. A phylogenetic tree of 57 PV types based on a neighbor-joining (Kimura two-parameter distances) evaluation of a 399-bp E6 gene alignment. The abbreviations are as defined in the legend to Fig. 1. FIG. 6. A histogram of simple distances (dissimilarities) between pairs of taxa for the 291-bp L1 segment, separated into within-group, across-group but within-supergroup, and across-supergroup comparisons. Note the overlap in the distributions of the distances, indicating the impossibility of using distance thresholds to unambiguously define the groupings of interest (see text for details). classification may also be defined on the basis of larger distance thresholds. However, we found that it is impossible to establish criteria for supratype classification of PVs that are based purely on distance (or its equivalent, similarity) and that give results which are compatible with reconstructions derived by phylogenetic methods or which are operationally consistent. The problem can be illustrated by the use of simple distances. As most nucleotide sequence distances between types of the same supergroup are less than 37% and those between types belonging to different supergroups are larger, one might consider defining a taxonomic category using this distance score as a cutoff point. When one examines the 291-bp L1 segments, one finds that HPV-33 and HPV-16 (both genital HPVs in supergroup A) are 24% dissimilar but that HPV-33 and HPV-38 (a cutaneous HPV in supergroup B) are 30% dissimilar and that HPV-16 and HPV-38 are 38% dissimilar. Given the 37% cutoff point, the distances from HPV-33 indicate that all three HPV types should be in the same supergroup, and yet given the pairwise distances of HPV-16 and HPV-38, these two types would have to be assigned to different supergroups. Similar conflicts can be found for many different types and at many different thresholds. Furthermore, this problem is not resolved under various distance corrections (data not shown). Figure 6 shows a distribution of the simple distances for intragroup, between-group, and between-supergroup (including the isolated types) comparisons for the specific taxonomic groupings presented in Table 2. It shows that while the modal distances are clearly different, there is a considerable overlap of the distributions such that no consistent set of classificatory thresholds can be nominated as an operational criterion. While this set of groupings does not give the minimal overlap of distances, the problems mentioned above preclude the definition of virtually any useful grouping on the basis of distances alone. Phylogenetic classification. Cladistic phylogenetic classifications reflect hypotheses about the evolutionary history of organisms (69, 70), and given the best hypothesis, all acceptable classifications must reflect monophyletic relationships (28). The same phylogeny may be used to establish alternative classifications which differ in the number of taxons and ranks, the decision being a matter of simplicity of use and consensus among taxonomic researchers. Table 2 reflects our proposal for a phylogenetic classification of the PVs which is aimed at minimizing category names while introducing supratype groupings at levels we believe will be biologically interesting. The choice of cladistic groups was guided by phenetic similarities in biology, disease pathology, and high bootstrap values; the results to a large part reflect traditional groupings that PV biologists have found useful. Occasionally, the distance criterion was used to determine the taxonomic organization, such as that between and within supergroups C to E. We have adopted informal taxon names such as group and supergroup, which are useful for taxa within families or genera (70), but have avoided the use of formal taxon names, including the names of specific viruses. The use of alphanumeric names maintains the stability of the nomenclature in the face of possible changes to the phylogeny and group membership because it keeps the group nomenclature independent of its constituent taxon names. By definition, a clade is formed by all species with shared derived ancestral characters and their most recent common ancestor. Because of the lack of a fossil record, viral phylogenies cannot be firmly rooted, and so we cannot be sure of the temporal direction of homologous character changes, which is required to recognize monophyly. We used the rooting of the weighted parsimony analysis (62) and possible outgroups such as the BPVs to root the HPVs, as suggested by the evidence of virus-host coadaptation with strict host specificity (6, 7, 56), to partially overcome this uncertainty. Remarks about many of the groupings have been addressed in the HPV database compendium (48), and additional points will be raised in the Discussion.

7 3080 CHAN ET AL. J. VIROL. DISCUSSION This paper investigates how nucleotide sequence information may be used to reconstruct the genealogies of PVs as a prerequisite for a phylogenetic taxonomy. The genomic segment that we chose meets the strict criteria for establishing phylogenies, is sufficiently discriminating for typing purposes, and has the advantage over other conserved genomic segments in being easily amplified by PCR. In the following, we describe our view of the taxonomy of PVs at the levels of variants, subtypes, types, groups, and supergroups, which correspond to increasing phenetic levels of diversity. For completeness, we have included in this discussion variants and subtypes, although the analysis below the type level was not a part of this study. Variants of PV types. Variants of a PV type differ from the reference type (the clone sequenced to describe that type) by distances of up to within genes and slightly more outside genes, a statement which reflects empirical observations rather than a definition. Sequence variants from HPV-5 and HPV-8 (11, 11a), HPV-6 and HPV-11 (27, 37), HPV-16 (19, 31, 32, 52), HPV-18, and HPV-45 (50) have been systematically studied. While the finding of sequence diversity among independent isolates is somewhat trivial, it is remarkable that no larger diversities were observed, i.e., the continuous process of substitutional alterations that has linked PV types in their evolutionary history has no complete record in today s HPV isolates. Old and new concepts about subtypes. The term subtype was originally used for certain HPV isolates of a type that had restriction patterns different in Southern blots from those of the reference type. Point mutations in restriction sites can alter such patterns dramatically, and it was thought that these isolates might differ significantly elsewhere in their genomes. Subsequent sequence analysis of subtypes of HPV-5 (71) and HPV-6 (27, 37) revealed a sequence diversity of only about 1%, i.e., as little as that between variants. The long control regions of independent isolates of HPV-6 and HPV-11 can differ by small insertions and deletions (27, 37) which are probably derived from sequences with a propensity for generating slippage (27), but these isolates do not differ significantly elsewhere in the genome. Against this background, we recommend that the term subtype be abandoned for these isolates. However, it can be useful for describing isolates with distances of to from their reference PV types, i.e., a diversity that is lower than that among types and higher than that among variants. Surprisingly, subtypes in this new sense of the word have been very rarely encountered, and we are aware of only three examples at the nucleotide level, the pairs HPV-34 HPV-64 and HPV-44 HPV-55 and the HPV genome in the ME180 cell line in comparison with that for HPV-68 (15, 54). PV types as species. The well-known biological species concept of Mayr (43, 44) applies exclusively to sexually reproducing species. Less well-known is the fact that it is a special case of a broader concept, the evolutionary species (45), which explicitly includes asexual organisms. We take the definition of an evolutionary species to be a single lineage of individuals separated from other extant lineages and from their common ancestors by substantial genetic differences (58, 69). Why is it important to identify the evolutionary species, and to which level of the organization of PVs does it correspond? The species is the highest taxonomic level (i.e., the largest group of individuals) subject to evolutionary selectional pressures through direct competition and as such the highest group that will tend to evolve as a unit. Higher taxa are mere groupings of species that exist solely by virtue of being named (43). We have previously implied that the PV type is the evolutionary species (3, 4). As such, the type identifies a population of viruses which may show minor internal variation and a large genetic discontinuity between types. PV speciation and the basis of classification from molecular sequences. Sexually replicating species achieve cohesiveness through a vertical bond between parents and offspring and a horizontal bond between mates. In the face of evolutionary novelties, cohesiveness of the species can be maintained, provided that mating bonds are not disrupted and gene flow occurs among mating populations. Asexual species lack the mating bond, and species cohesiveness depends on a mixture of stabilizing selection, stable environments, and genetic cohesiveness, whose molecular basis is the high fidelity of DNA replication and the low frequency of mutations (55). While evolutionary processes in sexual and asexual lineages undoubtedly differ, both reproductive modes result in lineages of organisms and hence are amenable to classification by the same principles (58). It is a characteristic of today s PVs, as with many other organisms, that one observes related genetic entities (the species or types) separated from each other by genetic discontinuities. This may in principle result from either separation of lineages into distinct populations via geographic segregation or adoption of different ecological niches or, if mutation rates are high enough and levels of competition between lineages are low enough, random mutation and drift of lineages which eventually accumulate high levels of diversity. As an ancestral type diverged into distinct lineages, intermediate variants would tend to become extinct, either because they were not as fit or simply as the result of chance, leaving behind relatively isolated lineages. The many host-specific PVs and the adaptation to cutaneous or mucosal epithelial environments support the role of selection for lineages specifically adapted to different ecological niches, but the number of types occupying similar ecological niches with similar geographic distributions suggests that mutation and random extinction of lineages must have played a part in the development of the current range of PV types as well. The operational definition of a PV type was arbitrary but apparently felicitous, insofar as few borderline cases and no genome less than 10% from both of two previously identified types have been confirmed. To some part, this observation can be explained on a statistical basis, because the evolution of sequences is, to a first approximation, governed by a molecular clock, so that mutations accumulate at roughly equal rates in different lineages. The consequence is that, in a system evolving in this fashion, distances back to a common ancestor will all be approximately equal, and if two sequences are more than a certain distance apart, the chance of a third sequence being less than that distance from both of the two previous sequences is vanishingly small. As this assumption seems to apply to PVs, it is likely that PV classifications at the type level will remain stable after future additions of novel genomes. As to the apparent scarcity of subtypes (defined as sequences between 2 and 10% different from a prototype sequence), it would be of interest to investigate whether the number of subtypes is less than would be expected under various models of evolution and population genetics. PV groups and supergroups. Below the taxonomic level of the subfamily, we have avoided the term genus, which might eventually become introduced after further study of PV genomic diversity. Instead, we use the neutral term supergroup, and in the following, specific characteristics of these supergroups and some isolated lineages will be addressed. Phenotypically, the recognition of supergroups A and B is supported by well-known differences in natural disease biology

8 VOL. 69, 1995 PV PHYLOGENY AND TAXONOMIC CLASSIFICATION 3081 and by gene regulation differences, like similar cis-responsive elements that are shared by types of the same supergroup (18, 63). However, both supergroups do contain some anomalies, which are discussed below. Supergroup A. All HPV types traditionally called genital and mucosal HPVs belong to supergroup A. Neither term is strictly correct, because typical genital HPVs (e.g., HPV-6) are also found in oral and laryngeal mucosas, and at genital sites, they also infect the cutaneous epithelia, e.g., the penile foreskin. Presently, it is not known whether any gene functions are unique to the genital HPV types. However, they do exhibit a common architecture in the long control region which is absent from unrelated PVs and which probably correlates with similar regulatory properties. Most conspicuously, there are an SP1 and two E2 binding sites immediately upstream of the TATA box that is involved in E6 transcription (63). Within supergroup A, we distinguish 11 groups on the basis of monophyly, high bootstrap scores (generally over 90%), and biological similarity. This classification highlights several interesting features of the supergroup s evolution; in particular, there are three groups associated with cutaneous lesions. Members of group A2 are normally found in skin warts, and group A4 viruses exhibit a dual predilection for both cutaneous and mucosal surfaces (4, 15). Group A8 is an unusual mixture of the cutaneous butcher s wart virus HPV-7 and two other mucosal types. These peculiar associations are also supported by nucleotide alignments of the E6, E1, and different parts of the L1 genes and therefore do seem to reflect natural evolutionary groupings (6, 7). The members of these groups may hold the secrets of how the mucosal-cutaneous distinction arose. The groups A6, A7, and A9 include the most commonly found high-risk types (4, 39), and these are the groups that have attracted the attention of most researchers. However, it is unfortunate that A7 and A9 have rather low bootstrap scores. As members of these groups are among the most widely studied, there is hope that a future synthesis of molecular functional data coupled with a comparative approach will help to make the relationships more clear. Most members of the remaining groups have been isolated from benign genital or oral lesions or from latent infections, but on occasion, they can also be found in malignant disease (15). Previously, hierarchical associations of genital HPVs reflecting similar biologies have been proposed (65). The cutaneousgenital HPVs (groups A2 and A4) were claimed to associate with the HPV-6 lineage to form the low-risk types, while most other HPVs formed the high-risk types. Our reexamination of the same E6 data set does not support this model. E6 phylogenies constructed by distance matrix, parsimony, and maximum-likelihood methods and evaluated by bootstrapping reveal that the cutaneous-genital HPVs, the low-risk genital HPVs (e.g., group A10), and the high-risk genital HPVs form independent groups within supergroup A. There is no reliable nested hierarchy of clades for these groups and consequently no evidence of a nested hierarchy of virulence. Supergroup B. EV is a rare skin disorder characterized by disseminated wartlike lesions and skin cancers (for reviews, see references 25 and 33). The 17 HPV types originally isolated from EV patients form the majority in supergroup B. Two groups, B1 and B2, are well-defined, and they both also contain types that have not been found in EV-associated lesions. Group B2 is more atypical, as only HPV-50 is considered to be an EV virus, while the rest have been isolated from non-ev patients and from such lesions as common warts, flat warts, and squamous cell carcinomas (for a review, see reference 15). Nevertheless, the grouping also has the support of amino acid sequence similarities throughout the entire genome (13). Group B1 is a large group of mainly EV HPVs, except for HPV-37, HPV-38, and HPV-49. The phylogenies are conflicting as to the internal group structure but do give a hint of subdivisions partially consistent with previous hybridization studies (38). In particular, there is agreement that HPV-9, HPV-15, HPV-17, HPV-22, HPV-23, HPV-37, and HPV-38 form a clade which also has a high bootstrap value of 93%, and the remaining types, excluding HPV-49, form another clade (bootstrap value, 91%). Because of the large distances between types in this group, it will be a prime candidate for splitting into smaller groups when the phylogenetic relationships are better understood. Recently, six new HPV genomes (2) have been detected by PCR in cutaneous cancers of renal transplant patients. A neighbor-joining analysis of a segment overlapping our 291-bp L1 region shows that they belong in group B1. Finally, it should be pointed out that EV patients frequently have flat warts associated with HPV-3 and HPV-10, two types that phylogeny assigns to the genital HPVs, although pathologically they are sometimes included among the EV HPVs. Supergroups C and D. Each of these two supergroups reliably unites ungulate PVs, which have always been considered as distinct because of their unique genomic organization and distinctive pathologies, i.e., fibropapillomas (BPV-1, BPV-2, BPV-5, European elk PV, deer PV, and ovine PV) versus true papillomas (BPV-3, BPV-4, and BPV-6) (for references, see reference 34). Supergroup E. In this supergroup are united the two fairly closely related viruses HPV-1 and HPV-63 (17), together with the very distantly related type HPV-41. On the level of very large distances, these three HPV types form a clade together with canine oral PV and cottontail rabbit PV. We have excluded the most distantly related Mastomys natalensis PV from this supergroup. Future animal PV sequences may better define these relationships. FPV. PVs appear to be widespread among vertebrate hosts other than mammals, but only one PV sequence from a bird, the chaffinch, has been reported (46). To estimate the relationship of FPV to other PVs, we aligned the published sequence against homologous segments of typical representatives of the five supergroups, i.e., HPV-1, HPV-5, HPV-16, BPV-1, BPV-4, and Mastomys natalensis PV (data not shown). Nearly all dissimilarities between the FPV sequence and those of the other six PVs exceeded those among the six PVs. We take this to indicate that in spite of the limited interpretability of simple distances, FPV is less related to any of the mammalian PVs than they are related to one another. Taxonomic flexibility. This taxonomic scheme is an outline and does not involve the full spectrum of activities traditionally associated with erecting a taxonomy, e.g., provision of identification keys. However, it does show how one may organize PVs by cladistic principles while giving recognition to levels of biomedical interest. It is an expression of our best phylogenetic hypotheses and should not be seen as an unchangeable convention but rather as a tool to inspire research to change it for the better. The proposed names do not mean that other collective names (e.g., high-risk group) are discouraged; the taxonomic convention permits these so long as they are useful and it is understood that they have no formal taxonomic status within a cladistic classification. Additional classificatory levels such as genus and subgroup may be erected when needed. We hope that an increasingly comparative approach aided by the HPV database (48) will speed us toward the goal of an increasingly informative taxonomy. Host-PV evolution. All known PVs have been detected in a single mammalian or bird host species (61), and no interspe-

9 3082 CHAN ET AL. J. VIROL. cies transfer has yet been reported. This host specificity and the inapparent or benign nature of most PV infections suggest that these viruses are extremely well-adapted parasites. We had previously suggested that this could indicate host-pv coevolution (7). A reexamination of the phylogeny and arguments derived from plant-insect studies (36) now indicates that it does not accord with coevolution in the strictest sense of the word (26, 35) for the following reasons. (i) The virus phylogeny does not mirror the host phylogeny accurately (Fahrenholz s rule), e.g., in supergroups A and C and in the dissimilarity between supergroups A and B compared with the other supergroups. (ii) In hosts harboring many different types of welladapted parasites, e.g., humans and cattle and their PVs, the selection pressure exerted by any one kind of parasite on the host is negligible by comparison with what the host can exert on the parasites. Thus, PV evolution is more likely to be dominated by unilateral host selection (sequential evolution [55]) rather than by mutual coevolution, as it has to adjust to the molecular mechanism of the host cell (56). The presence of three monkey PVs closely related to HPVs of supergroup A is an interesting observation that may be explained as a consequence of infection of monkey-human ancestors by the PV ancestors of supergroup A. Given that HPV-16 and HPV-18 variants indicate a very slow rate of evolution (as little as one mutation in 300 bp over several thousands of years) and assuming a molecular clock (72), supergroup A radiation could have overlapped that of its primate hosts. Alternatively, interpretation of nonsynonymous- to synonymous-change plots, especially with reference to the HPV- 13 pigmy chimpanzee PV comparison, suggests that the closeness could also be the result of cross-species transmission (48). PV evolution would then have parallels to the evolutionary history of the human immunodeficiency virus-simian immunodeficiency virus group (47). Although reports are lacking, it may be possible to test for PV viability and gene function following experimental infection of heterologous hosts. Should there be no cross-species viability, one might then attribute an even greater age to the affinities between animal PVs and HPVs in supergroup E and all other PVs, an age possibly similar to that of land vertebrates, i.e., about 200 million years. But whatever the opinions, there is certainly a need for the sampling and study of many more animal PVs. Future directions. The establishment of the HPV sequence database at the National Laboratory in Los Alamos by the National Institutes of Health (48) should give a boost to comparative studies. Research done there will be published annually in the form of a compendium that will be sent to interested scientists and that will provide updates on the relationships of newly typed PVs. One interesting project will be a search for correlations between amino acid and nucleotide sequence elements common to subgroups of PVs which may shed light on the molecular basis of PV biology and pathology. In the search for novel HPVs, one probably should not expect many more genital PVs, as most of the members of this group may have been found (5). However, the present phylogenetic isolation of HPV-1, HPV-41, and HPV-63 may indicate that many cutaneous HPV types (17, 30) are still to be found, and the recent identification of new EV-related viruses (2, 57) seems to prove this. Aside from this, it is apparent that future progress in the understanding of PV evolution will require the discovery of large numbers of novel animal PV genomes, e.g., genital PVs from primates or nonprimate mammals. On a potentially even deeper level of phylogenetic branching, it would be of interest to examine the relationship between PVs and polyomaviruses, which form subfamilies within the family Papovaviridae. They had been grouped together because they are the only viruses with double-stranded, covalently closed, circular DNA and with similar capsid morphologies. A monophyletic origin is rather doubtful because of the vast differences in genome organization and lack of sequence similarities. Only two short amino acid sequence similarities, which may be homologous, are known: (i) the binding site for the retinoblastoma protein of the simian virus 40 T antigen and of some HPV E7 proteins and (ii) a segment of the simian virus 40 T antigen that can be aligned with PV E1 proteins (8). These similarities possibly emerged through convergence, but even if they did reflect a common ancestry of protein domains, they may be taxonomically irrelevant, as they may be ancestral characters shared with eukaryotic or adenovirus proteins (12, 51). Evidence for shared derived characters could come from studies of three-dimensional structures. For example, one could speculate that the PV capsid proteins may have domains that interact with one another and with DNA in a manner similar to that of the domains of polyomavirus capsid proteins or that the E1 and E2 proteins might have DNA binding domains homologous to those of the polyoma T antigen. But on the strength of the present evidence, polyomaviruses and PVs should be considered separate families. ACKNOWLEDGMENTS We thank Shih-Ping Chan, Department of Mathematics, National University of Singapore, for computer assistance; Chiew-Hoon Tan for technical support; Gerald Myers, Theoretical Division, Los Alamos National Laboratory, for enlightening discussions; and the two reviewers for their invaluable suggestions. A.L.H. is a postdoctoral fellow sponsored by the Division of Microbiology and Infectious Diseases of the National Institute of Allergy and Infectious Diseases through an interagency agreement with the Los Alamos National Laboratory. REFERENCES 1. Banks, L., and G. Matlashewski Cell transformation and the HPV E5 gene. Papillomavirus Rep. 4: Berkhout, R. J. M., L. M. Tieben, H. L. Smits, J. N. Bouwes Bavinck, B. J. Vermeer, and J. ter Schegget Nested PCR approach for detection and typing of epidermodysplasia verruciformis-associated human papillomavirus types in cutaneous cancers from renal transplant recipients. J. Clin. Microbiol. 33: Bernard, H. U Coevolution of papillomaviruses with human populations. Trends Microbiol. 2: Bernard, H. U., S. Y. Chan, and H. Delius Evolution of papillomaviruses. Curr. Top. Microbiol. Immunol. 186: Bernard, H. U., S. Y. Chan, M. M. Manos, C. K. Ong, L. L. Villa, H. Delius, C. L. Peyton, H. M. Bauer, and C. M. Wheeler Identification and assessment of known and novel human papillomaviruses by PCR amplification, restriction fragment length polymorphism, nucleotide sequence, and phylogenetic algorithms. J. Infect. Dis. 170: Chan, S.-Y Ph.D. thesis. National University of Singapore, Singapore. 7. Chan, S.-Y., H.-U. Bernard, C.-K. Ong, S.-P. Chan, B. Hofmann, and H. Delius Phylogenetic analysis of 48 papillomavirus types and 28 subtypes and variants: a showcase for the molecular evolution of DNA viruses. J. Virol. 66: Clertant, P., and I. Seif A common function of polyoma virus large-t and papillomavirus E1 proteins. Nature (London) 311: Coggin, J. R., and H. zur Hausen Workshop on papillomaviruses and cancer. Cancer Res. 39: Danos, O., and M. Yaniv E6 and E7 gene products evolved by amplification of a 33-amino-acid peptide with a potential nucleic-acid-binding structure, p In B. M. Steinberg, J. L. Brandsma, and L. B. Taichman (ed.), Papillomaviruses (cancer cells 5). Cold Spring Harbor Laboratory, Cold Spring Harbor, N.Y. 11. Deau, M.-C., M. Favre, S. Jablonska, L.-A. Rueda, and G. Orth Genetic heterogeneity of oncogenic human papillomavirus type 5 (HPV5) and phylogeny of HPV5 variants associated with epidermodysplasia verruciformis. J. Clin. Microbiol. 31: a.Deau, M. C., M. Favre, and G. Orth Genetic heterogeneity among human papillomaviruses (HPV) associated with epidermodysplasia verruciformis: evidence for multiple allelic forms of HPV-5 and HPV-8 E6 genes. Virology 184: Defeo-Jones, D., P. S. Huang, R. E. Jones, K. M. Haskell, G. A. Vuocolo,